Blooms of Alexandrium occur in the Gulf of Maine each year and produce toxins that can accumulate in shellfish, causing Paralytic Shellfish Poisoning. Regional management agencies conduct rigorous monitoring to ensure the safety of shellfish consumers and support the shellfish industry. The project will implement a machine learning-based forecast with high spatial and temporal resolution to predict paralytic shellfish toxins in specific locations one to two weeks before shellfish become dangerous to eat.
Why We Care
Paralytic shellfish toxins (PSTs) are produced by several species of phytoplankton and can accumulate in shellfish and present a potentially lethal threat to consumers. Extensive blooms of the PST-producing organism, Alexandrium catenella, occur in the Gulf of Maine. Maine, New Hampshire, and Massachusetts use considerable resources to conduct rigorous monitoring to ensure the safety of shellfish consumers and support shellfish producers. Each summer, shellfish beds are closed due to dangerous levels of PSTs along large stretches of the coast. The shellfish aquaculture and harvesting industries, vital components of a sustainable waterfront economy, are severely disrupted by these closures.
A Gulf of Maine Harmful Algal Bloom stakeholders’ meeting in March 2018 identified the need for targeted forecasts to better predict the onset and decline of shellfish toxicity in specific regions and bays over timescales ranging from days to weeks before shellfish become dangerous to consumers. To address this need, researchers developed a highly accurate machine learning-based forecast using chemical “fingerprints” of PSTs in Maine shellfish. The neural network was trained on PSTs determined in Maine shellfish collected from 2014 to 2016, and predicted 2017 PST levels at individual locations one to two weeks in advance with over 95% accuracy.
What Are We Doing
The researchers are using a chemical analytical approach, high performance liquid chromatography (HPLC)-based PST toxicity analysis, to determine concentrations of twelve individual compounds that make up total PST levels. The relative composition of these compounds varies seasonally and geographically and these patterns of variation are captured in the machine learning analysis to generate predictions of total PST levels. Maine Department of Marine Resources (DMR), in collaboration with Bigelow Laboratory for Ocean Sciences, is the first and only state agency to use the HPLC method for shellfish safety in lieu of the mouse bioassay that determines total toxicity.
This project will test and refine the machine learning predictive capacity by incorporating the forecast as a product of the routine shellfish PST monitoring programs in the Gulf of Maine. An automated data pipeline will generate a machine learning-based PST forecast for 30+ specific locations along the Maine coastline from HPLC-based PST monitoring data. The information for each location, including PST levels in previous years and the current shellfish toxicity forecast, will be available on a web-based platform and in customized written reports to end-users. During the 3-year project, the value of the forecast and communication tools will be evaluated and improved using real-life applications by state regulatory managers at Maine DMR and shellfish producers and wholesalers. The project will also explore the value of extending the Gulf of Maine coastal forecast to New Hampshire and Massachusetts coastlines by initiating development of the relevant datasets and testing forecast accuracy.
Dr. Stephen Archer of Bigelow Laboratory for Ocean Sciences leads this project. Co-investigators are Dr. Nicholas Record (Bigelow Laboratory for Ocean Sciences), Ms. J. Kohl Kanwit (Maine Department of Marine Resources), and Dr. Jill MacLeod (Maine Department of Marine Resources).
The project is funded through the NCCOS Monitoring and Event Response for Harmful Algal Blooms (MERHAB) Program.